Is Deer Movement affected be Recreational Activity?

Submission for the MSc Module Computational Movement Analysis

To does

• Cross references

• Make plots larger

• Show random points similar to trajectories

Abstract

Findings in short.

1 Introduction

In addition to being the main habitat of European roe deer (Capreolus capreolus), forests provide diverse and outstanding settings for recreational activities such as hiking, horse riding and mountain biking (Coppes et al. 2017)

The following hypotheses are therefore being tested:

Hypotheses

Null hypothesis	Alternate hypothesis
Deer movement does not vary between weekdays and weekend	Deer movement on average increases during the weekend
Deer movement is not constrained by recreational activity	Deer movement is constrained by activity
There is no relationship between the popularity of recreational paths and deer activitiy	Deer movement is influenced by the popularity of recreational paths

2 Material and Methods

2.1 Datasets

The deer dataset was provided from a research project conducted in partnership with the Wildnispark Sihlwald. The Wildnispark Zurich consists of a protected core area and a surrounding area with high recreational use due to its proximity to the city of Zurich. The dataset tracks the movement of 14 deer using GPS trackers. Each fix point in the dataset contains the following attributes:

Data structure

Name	Purpose
Latitude	Positioning data
Longitude	Positioning data
x	Positioning data (projected)
y	Positioning data (projected)
reh	ID of individual deer
datetime_utc	Date and Timestamp

The characteristics of the data limit the computational movement analysis methods that can be applied.

Col1	Col2	Col3	Col4

To examine recreational activities additional data from the fitness tracking platform Strava was gathered by scraping the publicly available global heatmap using a script developed by Nils Ratnaweera. The extracted raster tiles contained colour values to indicate the number of activities. This provided a relative indication of how often a route was frequented. All activity types were (cycling, running etc.) were included and the dataset contained no temporal data. Lastly street data in the publicly available swissTLM3d dataset provided by swisstopo was used to examine the influence of recreational paths.

2.2 Preprocessing and Exploratory Analysis

The provided deer data was re-projected to LV95 and the Euclidean distance and time lag were calculated to perform further movement analysis. Based on the spatial extent of the entire dataset a bounding polygon was generated and used to crop Strava activity data and recreational path data.

The deer dataset consists of 14 individual deer, tracked over differing periods between 2013 and 2016 (Figure 1).

Figure 1: Numer of deer and respective sampling periods.

Multiple sampling regimes, in intervals of 180 and 5 minutes were used Figure 2 . The shorter sampling periods were applied inconsistently and had a short average duration of 6 hours. The longer intervals of 180 minutes was therefore preferred to examine the long-term relationships between deer movement and recreational activity. Temporal outliers were removed by filtering to ± 2 minutes of the long sampling interval.

(a) Timelag distribution

(b) Distance distribution

Figure 2: Frequency plots

Spatial extent

Spatial outliers were roughly removed by filtering the data to a step length of less than 2500 meters.

Due to the large number of individual deer temporal extent a subset of deer (RE08, RE11, RE06) was chosen due to their relatively high concurrent activity levels (average distance moved between measurements), spatial proximity to recreational infrastructure and spatial independence from one another. The multi-annual sampling period was also reduced to examining 2014, as there was continuous and well-sampled data for the chosen deer groups. This was further reduced to examining the months of January, April, July and October.

(a) Spatial distribution of selected deer

(b) Sampling frequency of selected deer

Figure 3: Deer subset chosen for further analysis

2.3 Methods

2.3.1 Influence of the weekend on average movement

Filtered deer movement was grouped by deer, month and whether movement was captured on a weekday or on a weekend. Average step length was calculated for each grouping and the values were compared using an ANOVA. A log transformation was applied to achieve a more normal distribution of the model residuals.

2.3.2 Influence of recreational paths

Two methods were used to examine the influence of recreational paths on deer movement. The first method used examined the number of times a deer trajectory crossed recreational paths and if there was any variation between weekdays and weekends.

(a) Deer 06

(b) Deer 08

(c) Deer 11

Figure 4: Deer movement in relation to hiking paths

The second method involved creating multiple buffers in 5m increments (5m, 10m, 15m, 20m, 25m) around recreational paths. For each examined deer and month the spatial extent and number of recorded points was used to generate a random distribution of points. Using an intersect the frequency of points within different buffers was compared to determine whether the relationship between deer and recreational paths follows a non-random pattern.

Figure 5: Observed values

Figure 6: Simulated values

2.3.3 Influence of popularity

The influence of path popularity was determined by extracting raster values from the strava heatmap and assigning them to the individual segments. A single 25m buffer around all street data was used to determine the number of deer recorded in the vicinity and a Spearman Rank correlation was used to determine whether path popularity and deer presence correlated.

(a) All streets

(b) Hiking paths

Figure 7: Popularity data from strava extracted to path data

3 Results and Discussion

3.1 Influence of the weekend on average movements

The ANOVA results ?@tbl-aov-results indicate that there is no significant difference in mean step length between weekdays and weekends. Whilst significant differences were observed between sample months and individual deer no significant interaction with whether data was recorded on the weekend was found.

?(caption)

term	df	sumsq	meansq	statistic	p.value
weekend	1	0.000	0.000	0.000	0.988
month	3	77.302	25.767	21.124	0.000
reh	2	49.837	24.919	20.428	0.000
weekend:month	3	2.855	0.952	0.780	0.505
weekend:reh	2	2.432	1.216	0.997	0.369
month:reh	6	32.118	5.353	4.388	0.000
weekend:month:reh	6	4.878	0.813	0.666	0.677
Residuals	2799	3414.276	1.220	NA	NA

Figure 8: Average step length between weekdays and weekend data

A further exploration using hourly averages for the same dataset showed a distinct pattern, matching the natural crepuscular movement of deer (Figure 9). Whilst data granularity was low seasonal shifts associated with shorter days in January were also observed. This implies that aggregated movement data is suitable for examining deer behaviour. However the influence of humans cannot be clearly identified at this scale. This could be a consequence of deer “freezing” in response to perturbations (Marantz et al. 2016), the duration of which does not lead to substantial differences in step length across 3 hour sampling intervals. To better identify the influence of recreational activity a higher temporal granularity, accelerometer data, coupled with either specific perturbation events or visitor data are required.

Figure 9: Average step length within a days

3.2 Influence of recreational paths

3.2.1 Comparing crossings on the weekend

When examining the number of crossings across the same subset of deer a clear difference in frequency between weekdays and weekends was found. This indicates that deer are less likely to cross recreational paths during the weekend. This pattern is more in line with observed deer behaviour from literature, such as freezing and avoidance (Coppes et al. 2017). Therefore intersection frequency is a preferable method when deer data has low temporal granularity and aggregated recreational activity is examined.

Figure 10: Number of path crossings, comparing weekend to weekdays

3.2.2 Vicinity to paths compared to randomly generated points

On average, fewer observations were recorded in the vicinity of paths when compared to randomly generated data. This indicates that in addition to reduced crossing of paths over the weekend deer on average avoid paths.

Figure 11: Comparison of deer distribution near paths to random points

When split by deer and month it is clear that this trend is avoidance behaviour is not completely consistent. This indicates that additional factors influence deer behaviour.

Figure 12: Comparison of deer distribution near paths to random points, subset by month and deer

3.3 Popularity

A visual investigation of RE06 (?@fig-popularity-re)was used to identify whether path popularity could be one of these additional factors. The distribution of points was clearly influenced by the more popular path. Whilst deer measurements were largely concentrated between two paths, the deer was more likely to cross the less frequented secondary paths.

Figure 13: Distribution of RE06 over sample period

Figure 14: Distribution of RE06 over sample period

However further investigation involving the previously utilised subset of deer indicated no correlation between path popularity and deer presence (Figure 15). It is important to consider that the derived popularity data is relative, contains no temporal component and does not differentiate activity types. Platforms that provide this additional data, can be a valuable and inexpensive source of information. They can be used alongside traditional infrared counters to better examine deer interactions over longer time frames.

Figure 15: Correlation between popularity and frequency of recorded deer within a 50m buffer of streets

4 Conclusion and future research

As outdoor recreational activities increase in popularity there is a continued need to better understand what the effects are on nearby wildlife. Derived movement data from three deer over the course of 2014 highlighted that recreational paths influence their movement patterns, resulting in greater avoidance overall and reduced crossing of the paths during the weekends. To determine responses to individual perturbation events a higher sampling rate, combined with accelerometer data is however required. Increased outdoor recreation also presents an opportunity to integrate activity data to better identify potential hotspots that may have a disproportionate influence on deer behaviour. A combination of temporally concurrent GPS-tracked activity data, verified by on the ground counters are required to further examine long-term interactions between deer and recreational activities.

References

Coppes, Joy, Friedrich Burghardt, Robert Hagen, Rudi Suchant, and Veronika Braunisch. 2017. “Human Recreation Affects Spatio-Temporal Habitat Use Patterns in Red Deer (Cervus Elaphus).” PLOS ONE 12 (5): e0175134. https://doi.org/10.1371/journal.pone.0175134.

Marantz, Sierra A., Jed A. Long, Stephen L. Webb, Kenneth L. Gee, Andrew R. Little, and Stephen Demarais. 2016. “Impacts of Human Hunting on Spatial Behavior of White-Tailed Deer (Odocoileus Virginianus).” Canadian Journal of Zoology 94 (12): 853–61. https://doi.org/10.1139/cjz-2016-0125.